CGI-Based Synthetic Data Generation and Detection Pipeline for Small Objects in Aerial Imagery

Small object detection in drone-captured imagery presents critical challenges across environmental monitoring, urban planning, and disaster management. Current approaches struggle with accurate landmark identification due to varying altitudes, diverse landscapes, and object scale variability. This research proposes an innovative two-stage object detection pipeline specifically designed for aerial imagery analysis. Our approach integrates advanced data generation techniques with a novel detection methodology. Experimental results demonstrate significant performance improvements: precision increased by 29.3% (from 0.616 to 0.796), recall improved by 24% (0.699 to 0.867), and mean Average Precision (mAP@0.5) enhanced by 23.2% (0.677 to 0.834) compared to traditional YOLO models. The pipeline combines Computer-Generated Imagery (CGI) for synthetic data creation with a two-stage detection approach. The first stage employs YOLOv9 for efficient region of interest localization, while the second stage utilizes a ResNet-based model for precise classification. By addressing challenges associated with small object detection, our methodology provides a scalable solution for automated landmark recognition in diverse aerial environments.